ideology and the growth of government - university of bristol · ideology and the growth of...
TRANSCRIPT
Ideology and the Growth of Government
Andrew Pickering James Rockey
Discussion Paper No. 07/599
October 2007
Department of Economics University of Bristol 8 Woodland Road Bristol BS8 1TN
Ideology and the Growth of Government
Andrew Pickering James Rockey
29th October 2007
Abstract
We analyze the impact of ideology on the size of government. In a simple model
the government sets redistribution and provision of public services according to the
preferences of the median voter, for whom private consumption is a necessity. Ideology
is de…ned on preferences for public services and the impact of ideology upon the size of
government increases with mean income. In empirical work ideology is measured using
data based on party manifestos. Much of the increases and divergence in government
size observed across OECD countries can be explained by the interaction of ideology
and mean income.
Keywords: ideology, Wagner’s law, size of government
JEL Codes: D72, H10
1 Introduction
Over the post-war years total government spending as a fraction of GDP increased in all
OECD countries. However, growth rates have di¤ered markedly. In 1960 "the size of the
public sector was pretty much the same in almost all of the Western developed economies"
(Henrekson and Lybeck, 1988): the OECD average stood at 27.2%. By the late 1990s
total government spending stood around 60% in Sweden and well above 50% in many other
countries of continental Europe and around 35% in Japan, Switzerland and the United States
(Persson and Tabellini, 2003). This divergence is remarkable not least because over this same
time period there was substantial convergence in overall economic development across OECD
members, for example measured in GDP per capita. In this paper we ask whether and how
cross-country di¤erences in ideology, as captured by the ideological positioning of the median
voter, can determine these phenomena. We argue theoretically for, and …nd empirically, a
sizable and signi…cant impact of ideology, and most importantly the interaction of ideology
and the level of GDP per capita. The observed growth, and divergence of government size
is largely explained by ideology, and its interaction with income.
The size of the state has a venerable history as a subject for academic discourse. Theo-
retical studies have o¤ered numerous explanations for the growth of government, although
to date empirical work has not been conclusive.1 Holsey and Borcherding (1997) make a
distinction between ’apolitical’ and ’political’ explanations. The former relate to issues per-
taining to the demand and supply of public services. The latter relate to explanations based
on the political power of particular members of the electorate and perhaps more malevolently
1The following review draws in part from excellent surveys by Holsey and Borcherding (1997) and Lybeck(1988).
1
the incentives facing politicians under alternative constitutional rules.
On the apolitical side Wagner (1893) proposed a "law of increasing state activity" which
is now known as Wagner’s law. Whilst researchers have proposed alternative de…nitions of
this law (e.g. see Peacock and Scott, 2000) the standard interpretation now seems to be
that the income elasticity of demand for publicly provided goods is greater than one (Holsey
and Borcherding, 1997). Peacock and Scott (2000) distinguish between ’traditional’ govern-
ment services such as defence and law and order, and ’newer’ functions such as education,
health and welfare services including employment insurance and state pensions. According
to Wagner’s law the demand for these services in aggregate increases more than one-for-one
with advances in economic development. However, and perhaps surprisingly, Holsey and
Borcherding …nd that the empirical support in favour of Wagner’s law is "mixed at best".
Nonetheless we take Wagner’s law seriously: One particular concern is that Wagner’s law
has not been formalized within an economic model - an omission which we correct in this
paper. In our model income elasticity is positive but diminishing as income rises, and im-
portantly is di¤erent across countries given di¤ering underlying parameters. In particular,
income elasticity is highly dependent upon ideology - and previous empirical work has all
but completely neglected this aspect in determining government size.
Whilst Wagner’s law focuses on the demand side, Baumol’s (1967) cost disease represents
a supply side explanation. In this theory the public sector su¤ers from cost disease because
by assumption only the private sector enjoys technological progress. As wages rise the
relative cost of the public sector must therefore increase and given price-inelastic demand for
government goods the public-sector expenditure share must increase. The literature provides
some evidence in support of aspects of Baumol’s hypothesis (see Holsey and Borcherding,
2
1997), though whilst this is a persuasive story we argue that it does not explain the full
variation in the observed growth of the state in the comparative context. For example,
the cost disease explanation would attribute the greater public sector share in Sweden to
higher relative costs of publicly provided goods in that country. However, the theory asserts
that whilst expenditure increases, the actual quantity of the public services delivered in
equilibrium falls. It is abundantly clear that the actual quantity of public provided services
in the shape of redistribution, healthcare etc. is signi…cantly larger in Scandinavia than
in the US so we conclude that the Baumol explanation cannot fully explain the observed
di¤erences: there is something else going on.
On the political side a seminal contribution is that of Meltzer and Richard (1981). The
role of government in theirmodel is solely to redistribute, and voters have distinct preferences
over the tax rate given their position in the before-tax income (i.e. productivity) distribution.
The poor (rich) prefer higher (lower) taxes in this one-dimensional policy setting and in
equilibrium the median voter prevails. Because income distributions in practice are always
right-skewed the median voter chooses a positive tax rate, though this is less than unity
because voters rationally anticipate the disincentive e¤ect of higher taxes. Their central
result therefore is that greater inequality as captured by the di¤erence between mean and
median income leads to bigger government. Unfortunately the Meltzer and Richard (1981)
hypothesis has not fared at all well when confronted with the data. An obvious example,
again, compares Scandinavian countries with quite equal income distributions, yet large
governments against the US with the reverse. Our view is that cross-country comparisons
of this sort are subject to omitted variable bias, and clearly there are important cultural
and ideology variables that need to be taken into account. It is also the case that Meltzer
3
and Richard understate the complexity of the activities of government. As already alluded
the government’s role in the economy is multi-faceted, including provision of services that
don’t …t neatly into the de…nition of pure redistribution. By including public goods, and
furthermore ideology in the shape of appetite for these public goods, as well as redistribution
we …nd a potential resolution of this empirical failing.
Recent work has focused on the impact of constitutional rules upon the size of govern-
ment. Theoretical work by Persson and Tabellini (1999, 2000), Lizzeri and Persico (2001)
and Milesi-Ferretti et al (2002) argues for a strong e¤ect of the electoral rule, and in par-
ticular whether voting is aggregated proportionally or according to majoritarian systems,
upon the composition of public spending. The theory also is suggestive that proportional
representation leads to larger overall government. Persson et al (2000) argue theoretically
that parliamentary as opposed to presidential government will lead to a larger State. Em-
pirical work by Persson and Tabellini (2003, 2004) …nds that di¤erences in government size
across countries (measured as an average through the 1990s) are signi…cantly explained by
these two constitutional rules.2 This represents a substantial body of work which we aim to
improve upon in two directions. Firstly, as pointed out by Acemoglu (2005), there is a poten-
tial problem of omitted variable bias, and indeed Acemoglu suggests "di¤erences in culture"
amongst other variables as an important underlying factor that potentially co-determines
the constitutional rule as well as the choice of government size. Gabel and Hix (2005) also
make this point, and indeed …nd that ideology3 is signi…cant in the cross-section. However,
unlike Gabel and Hix we construct a model of the demand for government and argue for an
2The bottom line in Persson and Tabellini (2004, p. 39) is that presidential regimes and majoritarianelections are found to each cut the size of government by about 5% of GDP.
3In their study ideology measures are constructed using data from the World Values Survey.
4
interaction of ideology and GDP per capita. Furthermore our empirical analysis uses panel
data, hence we can control for country-speci…c factors that are …xed through time. Secondly,
and following on from this, we also note that Persson and Tabellini’s underlying theories are
essentially static: the predictions for government size hold regardless of the state of devel-
opment (i.e. the level of GDP per capita). However Persson and Tabellini (2004) report (in
column 5 of table 2) that there is no evidence of any constitutional e¤ects using data from
the 1960s. By allowing for a positive income elasticity of demand for public services, that
varies with ideology we derive and estimate a model that can explain the observations of the
60s as well as those of the 90s.
In sum there are a great many well-articulated theories explaining the growth of gov-
ernment. However, empirical work has been far from conclusive. We believe the elephant
in the room is the impact of tastes, culture or ideology.4 Indeed Holsey and Borcherding
(1997) write that "few deny that ideology matters in the size and composition of government
spending" (ibid, p. 587).
In the theory below our interpretation of ideology is that it represents preferences for
publicly provided goods and services, broadly de…ned: leftist cultures are keener on state
provision, and especially of Wagnerian newer functions such as health, education, social in-
surance and indeed what might be termed altruistic redistribution. Rightist cultures are
keener on provision by the market.5 However, and importantly, these preferences are tem-
pered by income. When incomes are low, the pain of taxation in terms of lost utility from
4An exception is Kau and Rubin (2002), who report that the Poole-Rosenthal measure of Senate ideologyhas a small impact on government revenues within the US.
5This is a reasonably consensual view of what is meant by "left" and "right". For example Gabel andHix (2005) write that "In general, voters on the left prefer higher taxes and higher levels of publc spendingthan voters on the right".
5
reduced private consumption is high. Even if the ideological climate is statist (i.e. left-
wing), the median voter in circumstances of low income will not tolerate big increases in
taxes. Consequently at low levels of income we would not expect to see much dispersion
across countries, and ideologies, in terms of government size. However, as income grows,
statist inclinations may be indulged. In terms of utility the pain of taxation is reduced at
higher levels of income and so the left-leaning median voter mandates higher taxes to fund
the Wagnerian goods she desires. The theory thus can predict what is observed in the data:
increases, and divergence in the size of government as GDP levels rise.
In our empirical work we employ the ideology measures of Budge et al (2001). These
data describe comparable ideological positions of political parties along a left-right dimen-
sion for most OECD countries over the period 1945-1998. Using these data Kim and Fording
(2001) derive estimates of the ideological position of the median voter in the country, and
…nd plausible results: Scandinavians are more ’left-wing’ on average, Anglo-Saxon countries
exhibited a right-ward drift through the 1980s and so on. A key advantage of these data is
that they vary through time as well as across space, allowing for analysis of how ideology has
a¤ected the size of government within as well as across countries. The approach taken here
is to regard ideology as exogenous. Fundamentally ideology in our context represents tastes,
and economics traditionally takes these as given. Nonetheless, as with everything, ideology
may well itself have its own deeper determinants and there is an extremely interesting liter-
ature exploring the related question of why some societies are more egalitarian than others.6
However, because we employ panel data our empirical analysis focuses on within-country
6For example Alesina et al (2001) attribute the absence of a welfare state in the US to racial heterogeneity.Alesina and Angeletos (2005) describe how redistributive policy changes depending on di¤erences in beliefson the extent to which income is determined by luck, birth and connections as opposed to merit.
6
variation, and so deeper cultural causes are controlled for in the …xed e¤ects. Furthermore,
because we would expect substantial lags between preferences and enacted policy the ideol-
ogy measure used in the empirical analysis is constructed as an average of lagged ideology
data. We are therefore con…dent that the relationship that we …nd between government size
and ideology, and especially the interaction of ideology and income is causal.
In the next section we derive the model. Section 3 contains our empirical work and
section 4 concludes.
2 The Model
The basic premise is that preferences for public goods and services, in other words ideology,
help to determine the size of government. In order to analyze this formally we modify the
political-economicmodel in the seminal paper by Meltzer and Richard (1981) (MR) to include
spending on a generic public good,7 de…ned as the Wagnerian components of government
spending. In MR the purpose of government is solely to redistribute, thereby …nancing
private consumption. The self-interested voter votes for redistribution so long as it exceeds
the lost consumption from taxation. However as argued above the functions of government
are much more diverse than just narrow redistribution hence in this paper taxation …nances
both redistribution and other public services. In fact it is conceivable that redistribution
desired for altruistic reasons may be thought more of as a Wagnerian public good rather
than redistribution in the MR sense. In order to derive an explicit solution for the size of
the government we follow Meltzer and Richard (1983) and employ the Stone-Geary utility
7Note that ’public goods’ here may be rivalrous and excludable thus are not necessarily pure-public goodsin the Samuelsonian sense.
7
function, which as they note is capable of showing whether the share of income taxed remains
constant, increases, or decreases as income changes.
The objective function therefore is written as
ui (ci, g) = ln (ci ¡ µ) + β ln (g +µ) (1)
where ci is consumption of person i and g is the per capita level of publicly provided goods
and services. As already discussed g represents a substantial component, and probably
the bulk of total government spending (although this is not necessary for our results to go
through). g represents the Wagnerian components of the government activity such as health,
education, and indeed part of redistribution, when it is demanded for insurance purposes
or even altruistic reasons. The public clearly likes these services, but they are separable
from private consumption in the utility function. β > 0 is a preference parameter that
re‡ects liking for publicly provided goods and services, which as argued above characterizes
society’s ideological position. The larger this is, the more left-wing the culture. Finally µ¸ 0is the Stone-Geary parameter. In the spirit of Wagner’s law we model private consumption
as a necessity and public goods as a luxury, so µ is interpreted as the subsistence level of
consumption.
Income, yi, di¤ers across individuals and is taxed at a linear rate, t. As in MR consump-
tion is augmented by lump-sum redistribution, r:
ci = (1¡ t) yi + r (2)
Taxation revenue …nances both redistribution and the public good and given a balanced
8
budget
tδ_y = g + r. (3)
where_y is mean income, and 0 < δ · 1 is a parameter capturing public sector e¢ciency
(high δ) or waste (low δ). This waste could be due to the argument that higher taxes reduce
incentives to work hence reducing mean income from what it would otherwise be, as well as
representing tax revenue expropriated or spent on non-productive ego-projects etc. De…ning
0 · ϕ · 1 as the share of tax revenue spent on public services, then
g = ϕtδ_y (4)
r = (1 ¡ϕ) tδ_y. (5)
The utility function can therefore be rewritten as
ui (ci, t) = ln¡yi (1 ¡ t) + (1¡ ϕ) tδ
_y ¡ µ¢+ β ln ¡ϕtδ_y + µ¢ . (6)
An assumption made here is that ϕ is …xed exogenously thus reducing the policy space to
one dimension.8 The gain in doing this is that, as shown by Roberts (1978) the political
equilibrium exists and is characterized by the choice of the median voter. It is well known
that a Condorcet winner generally will not exist given two policy variables. A possible
justi…cation for making this assumption could be public sector inertia: if various departments
of government have particular power in claiming shares of the overall tax revenue pot then
changes in relative spending may at least be slow. Voters would therefore take this as given
8An alternative would be to set ϕ = 1, which would be a special case of the model analysed.
9
when identifying their preferred tax rate. It can be seen from (6) that when ϕ = 0 the model
reduces to the case considered by Meltzer and Richard. In this environment the poor are
increasingly keen to tax and redistribute. In the case of ϕ = 1 we have a purely Wagnerian
model, and in this case it is the rich that are keener to tax given the utility function (note
either way and for all points in between the median voter is still the Condorcet winner).
By considering the general case we can identify whether and how the spending pattern of
government alters the impact of ideology upon government size.
The optimal choice of the median voter is given by di¤erentiating (6) and solving for t
giving
t¤ =βϕδ
_y ¡ [1¡ (1 ¡ ϕ) δm]µ¡ βϕδmµϕδ
_y (1 + β) [1 ¡ (1¡ ϕ) δm] (7)
where m =_y/ym > 1 is the ratio of mean income to median income (ym). In order to ensure
0 · t¤ < 1 two conditions are formally required:
_y ¸ mµ+ [1¡ (1¡ ϕ)δm]µ
βϕδ(8)
[1 ¡ (1¡ ϕ) δm] > β
1 + β(9)
Condition (8) ensures that t¤ ¸ 0 and can be described as a minimum development require-ment: mean income must be su¢ciently large relative to subsistence (µ) such that there
are incentives for the median voter to tax at a positive rate. Since our focus is on OECD
countries we argue this is a reasonable assumption. Condition (9) ensures that t¤ < 1. This
will obtain as long as the level of waste is high (i.e. δ is quantitatively small) enough to
dissuade the median voter from increasing taxes further. Given that it is plausible that δ
10
would approach zero as t approaches 1 due to the increasing disincentive e¤ect of taxation
this condition would be satis…ed for any value of β. Formalizing the exact relationship be-
tween δ and t substantially complicates the model, and more importantly is unnecessary for
our objective.
Given equation (7) and conditions (8) and (9) we have the following proposition, for
which formal details are available in the appendix.
Proposition 1 (i) Ideology. Holding_y, m and the other model parameters constant, a
higher value of β entails a larger state. The more left-wing a country’s ideology, the larger
its government.
(ii) Wagner’s law. Holding β, m and the other model parameters constant,
government grows with mean income. Furthermore the income elasticity of the demand for
government is positive, but diminishing.
(iii) Interaction. Given δm > 1 di¤erences in government size due to ide-
ology will increase with income.
Part (i) of proposition 1 is straightforward and unsurprising: given our de…nition of Left-
wing ideology as an increased liking for state provision, then a higher value of β corresponds
to relatively left-wing preferences, and a larger state, holding other variables constant.
Part (ii) of proposition 2 formalizes Wagner’s law. From ’take-o¤’ GDP,_y0 (i.e. that
level of_y de…ning (8) with equality) taxes increase from zero to some limit. Nonetheless in
this set up policy does not su¤er from Peacock and Scott’s (2000) ’reductio ad absurdum’,
11
that the state will eventually absorb the entire economy. From (8) and (7) we have,
_y0 = mµ+
[1 ¡ (1¡ ϕ) δm]µβϕδ
(10)
lim_y!1
t =β
(1 + β) [1¡ (1¡ ϕ)δm] . (11)
Equation (10) de…nes the level of development at which the state takes o¤. Equation (11) is
less than unity given (9). In a world of positive economic growth at the limit the public sector
converges to some ’steady State’ which is increasing in left-wing ideology and government
e¢ciency (i.e. characterized by high values of δ). The ’steady State’ also is increasing in
inequality (m) for precisely the reasons given by Meltzer and Richard (1981) - the further
away the median from mean income, the greater the extent of redistribution from the rich
to the poor, all else equal.
The income elasticity of the demand for overall government size is positive, although
diminishing as the economy grows. The fact that the growth of government is at …rst rapid
is due to the functional form of the utility function: once subsistence has been achieved, the
median voter then allocates a constant proportion of her marginal income to both private
consumption and the public good, g. Overall the proportion of total income devoted to
public sector activities (redistribution and provision of public goods) increases and converges
towards some level that is in large part determined by tastes for public provision or ideology
as characterized by the parameter β.
Finally, part (iii) of proposition 1 establishes that the impact of ideology upon government
size is state-dependent. If δm > 1 then the di¤erence in government size between two
countries of alternative ideological persuasion will expand as their economies grow, otherwise
12
the di¤erence will fall. We argue this condition is likely to be met within OECD countries:
as long as waste is not too serious (i.e. δ not too low), for a su¢ciently skewed income
distribution the impact of given ideological di¤erences increases as mean income increases.
Thus, di¤erences in government size in leftist and rightist countries at low levels of income
tend to be small. As incomes rise the capacity for state expansion rises and di¤erences in
ideology manifest themselves more concretely.
In …gure 1 we calibrate the model for archetypal left- and right-leaning countries (think of
Sweden vs the US). The model parameters are enumerated as follows: µ = $2000, ϕ = 0.75,
m = 2, δ = 0.8 (though the general argument is robust to alternative parameterizations). For
the left- (right-) wing country we parameterize ideology as βL = 0.6 (βR = 0.3). As would
be expected take-o¤ happens somewhat sooner in the left-wing country (at a mean income
of $7333 instead of $10667), although given the reasonably large di¤erence in ideologies the
di¤erence is not that vast. For example if the left-wing country (again think Sweden) has
a larger subsistence requirement (e.g. food & heating) then this could shift take-o¤ to the
extent it happens earlier in the right-wing country. Numerically if µ is increased to $3000
then take-o¤ in the left-wing country is put back to $11000. However, once income starts to
grow then the State in the left-wing country grows faster (because δm = 1.6 > 1) and the
ideological di¤erence between the two countries manifests itself increasingly strongly. At the
limit the left-wing country’s government grows to 62.5% whereas in the right-wing country
the state share of GDP grows to no more than 38.5%. It is this interaction between GDP
levels and ideology that we argue explains much of the growth, and divergence across OECD
countries in the latter half of the last century.
En passant we revisit Meltzer and Richard’s line of enquiry and examine the relationship
13
between government size and inequality. Di¤erentiating (7) with respect to m yields
∂t¤
∂m= ¡ β
©µ¡ (1 ¡ ϕ) δ_yª
_y (1 + β) [1¡ (1 ¡ ϕ) δm]2 ? 0. (12)
According to equation (12) the size of the state may be increasing or decreasing with respect
to inequality. The condition for this is whether or not (1¡ ϕ) δ_y ¡ µ is greater or less
than zero. The Meltzer and Richard case corresponds to ϕ = 0 and δ = 1 hence (12)
is positive for all of the relevant income range. In the general setting this holds once_y
becomes su¢ciently large and as apparent from equation (11) taxation must eventually be
increasing in inequality. Given positive economic growth eventually the Meltzer and Richard
hypothesis that inequality leads to larger government, will preside. However, at low levels of
GDP it is feasible that the median voter’s preferences are for lower taxes, given a su¢ciently
wasteful government (small δ) or a su¢cient weight on public goods in the budget (large ϕ).
Marginal utility of consumption is high at low income levels, and increased taxes only come
back partially as …nance for additional consumption due to spending on public services and
waste. We make these observations because empirical evidence on the Meltzer and Richard
hypothesis has on the whole been quite weak. The argument presented here re…nes their
argument by saying the relationship between inequality and government size is potentially
ambiguous at low levels of income when concerns over subsistence may dominate concerns
over (partial) redistribution in circumstances where the government spends money outside of
redistribution on Wagnerian public services and indeed may also be wasteful. In …gure 2 we
consider how the size of government evolves under equal (m = 1.5) and unequal (m = 2.5)
income distributions9 and …nd that at low levels of income, there is barely any di¤erence
9Here β is set equal to 0.5 and the other parameters are set as for …gure 1.
14
between the two. Indeed at mean income levels below $10,000 the preferred tax rate is higher
for the equal distribution. This is because at these low income levels the pivotal voter is
poorer under the unequal income distribution than the egalitarian income distribution, and
is more concerned with subsistence than the fruits of public spending.
Before proceeding to the empirical work it is worth considering how robust the central
proposition is to relaxing the assumption of exogeneity of ϕ. One possible conjecture would
be that ϕ increases with mean income (again holding the distribution of income constant).
The electorate, as it grows richer would have more of the ’luxury’ aspects of state provision
than pure redistribution, as the marginal utility of consumption would surely be falling.
Given a higher value of ϕ as income rises, then the increased impact of ideology is likely to
be further increased as g becomes a larger fraction of government spending.
3 Evidence
The empirical analysis focuses on 17OECD countries that have been democracies throughout
the post-war era. The countries are Australia, Austria, Belgium, Canada, Denmark, Finland,
France, Germany, Iceland, Ireland, Italy, the Netherlands, Norway, Sweden, Switzerland, the
UK and the US.10
The dependent variable is total government outlays as a percentage share of GDP, from
theOECD EconomicOutlook database. Figure 3 depicts these data, which show considerable
anti-cyclicality as would be expected, and more relevantly over the long run an upward trend
in all countries. The increasing dispersion across countries is also noteworthy, especially
10These countries were chosen purely on the criteria of their being democracies throughout the period andthere being available data for both total government outlays and ideology.
15
so given the convergence in GDP per capita and more broad development terms across
OECD members over this time horizon. Our argument is that the impact of ideology on
the median voter’s preference for public goods increases as marginal utility of consumption
of private goods declines. The median voter thus becomes increasingly liberated to exercise
her preferred public-sector outcome with increases in the level of economic development.
In …gures 4-6 the dependent variable is graphed against GDP per capita in constant 1985
dollars for each country. These graphs make clear a number of facts central to our argument.
Firstly over the sample there is a positive relationship between government size and GDP per
capita in all countries. Second, the relationship often appears to be non-linear. In second-
order polynomial trend regressions the linear term is positive, whilst the quadratic term is
negative in most cases.11 The only countries for which the quadratic term is positive are
Finland and France and in both cases these estimates are statistically insigni…cant: Given
the small sample size this provides reasonable support for the argument made above, that the
income elasticity of the demand for government is positive, but diminishing. Third, income
elasticity di¤ers substantially across countries. For example the Scandinavian countries and
some of the countries of mainland Europe exhibit fairly steep slopes, whereas in countries like
Australia, Iceland, Switzerland and the US government size is apparently less income elastic.
In recent years government size has fallen to an extent in a number of countries (Ireland, the
Netherlands and the UK are notable cases). This may be attributable to cyclical factors,
though might also be due to a rightward shift observed in the ideology data during the 1990s.
For our ideology data we utilize theManifesto Research Group (MRG) data of Budge et al
(2001). These data summarize left-right ideology at the level of the party, and following Kim
11This …nding is robust to the inclusion of the control variables as used in the regression analysis below.
16
and Fording (2001) annual series are constructed for the median voter ideological position
in each country. Country level averages of these data are presented in …gure 7 showing,
as would be expected, that the Scandinavian countries are on average substantially more
left-wing than say the US or Australia. It is perhaps surprising that the UK is left of the
Netherlands and Denmark on average through this time period, but as would be expected
the UK exhibits a marked drift to the right in the 1980s and the period 1945-1979 (i.e. prior
to Thatcher) constitutes around 2/3 of the data set.
Figure 8 illustrates the ideology series over time averaging over countries with Propor-
tional Representation (PR) and Majoritarian electoral rules respectively. Following Persson
and Tabellini’s (2003) de…nition the majoritarian countries are Australia, Canada, France,
the UK and the US. In the 1950s and 1960s both sets of democracies if anything exhibit a
trend to the left, though during this period neither is consistently di¤erent from the other.
Things change markedly in the late 1970s with the right-ward shift in the Anglo-Saxon coun-
tries. In the 1990s median voters in the PR countries have also apparently shifted to the
right, though they remain on average left of their colleagues in the Majoritarian countries.
Because in reality there are substantial lags between preferences as expressed in the ide-
ology data, and policy enacted by government, in the regression analysis below we use a
moving average of the previous 10-years’ MRG data for each country to construct a measure
of the current aggregate ideological position for that country. This procedure also miti-
gates against concerns of endogeneity - the ideology measures now substantially predate the
observations on government size, and are thus pre-determined. This measure is then trans-
formed12 to facilitate inference, so that our …nal ideology measure, ideoit are scaled from
12The MRG median voter data takes values from -100 (extreme left) to +100 (extreme right). We transform
17
minus unity (extreme right) to unity (extreme left). In …gure 9 averages of these data are
plotted against average government size in the 1990s. The correlation coe¢cient between
these two variables is 0.7, and the …gure shows that ideology can potentially explain a lot of
the observed variation in government size across countries.
To examine the relationship between government size, income and ideology more system-
atically we now turn to panel data analysis. The econometric analysis follows Persson and
Tabellini (2003) (PT), except the analysis here is restricted to just the OECD countries for
which we have ideology data and which have been democracies throughout. Their analysis
(out of necessity given their wider scope) focused on central government expenditure only.
However, for the OECD countries it is possible to use total government outlays as the depen-
dent variable (following the notation used in section 2 denoted t), which is preferable in that
it includes expenditures made by local government. We employ the same control variables13
used as standard by PT, and in addition, use their cyclical control variables14, as these are
in most cases signi…cant in the case of the OECD countries. Finally, the lagged dependent
variable is also included as a regressor throughout because of the substantial persistence in
this variable.
In table 1 total government outlays (t) are regressed upon income, ideology and the
interaction of ideology and income. Columns (1a) and (2a) show the results for the full
the data according to ideoit = (¡______
MRG it)/100 where______
MRG itis the mean of the past 10 years medianvoter position constructed as in Kim and Fording (2001).13Speci…cally ’PROP1564’ - the percentage of the country’s population between 15 and 64 year of age in
the total population, ’PROP65’ - the percentage of the country’s population over the age of 65 in the totalpopulation, ’TRADE’ - the sum of exports and imports measured as a share of GDP,14The cyclical control variables are ’YGAP’ - the deviation of aggregate output from its trend value in
percent, ’OIL_EX’ - the oil price multiplied by a dummy variable equal to 1 if net exports of oil are positiveand ’OIL_IM’ - the oil price multiplied by a dummy variable equal to 1 if net exports of oil are negative.
18
sample using …xed e¤ects but without common time e¤ects. Given the presence of the
lagged dependent variable the parameter estimates in these columns re‡ect the current-
period (or short-run) impact upon t of the explanatory variables. In columns (1b) and (2b)
corresponding ’long-run’ parameters estimates are presented,15 illustrating the impact of
particular levels of income and ideology upon the long-run steady state level of t - which we
denote t¤.
In column (1) the direct impact of income upon government size is positive and statisti-
cally and economically signi…cant. In this sample a one standard deviation ($3300) increase
in Y would increase outlays by about 4.4% of GDP in the long run. Perhaps surprisingly the
direct impact of ideology in column (1) is estimated to be negative - contrary to expectation.
However, the overall impact of ideology turns out to be positive because of the interactive
e¤ect with income. The full impact of ideology is modelled as a linear function of income,
and what we term here the direct impact is the intercept of that function. Because the slope
(interaction) term is always su¢ciently large, the full impact of ideology is positive as would
be expected.16
To assess the impact of ideology through time consider the increase in the size of the
State in what we term an archetypal left-wing country (say Sweden) compared with that in
15Given the regression tt = αtt¡1 + βYt + γIt + δYtIt + ... the long-run level of t is taken as t¤ =β
1¡α Yt +γ
1¡α It +δ
1¡αYtIt + ... = λYt + µIt + νYtIt + .... The long-run parameters, λ, µ, and ν and theirstandard errors are estimated iteratively using the Delta method.16Nonetheless, the negative direct e¤ect, even if in economic terms quite small (the standard deviation of
the ideology variable is 0.12) is an interesting …nding. One possible rationale could be that electorates areinclined towards left-wing regimes when the economy is on the upslope of the business cycle, but inclinedtowards right-wing regimes after the economy has peaked. Given the anti-cyclical nature of t this would leadto the negative direct relationship observed in the data. However, and most importantly, the interactionterm itself is estimated to be positive and statistically signi…cant at the 1% level, and is economically sizeableas well.
19
an archetypal right-wing country (say the US). The left- (right-) wing country has constant
ideology set equal to one standard deviation (0.12) above (below) the mean ideology score
(0.04) in the sample. In 1960 mean income was $6185, and in 1998 means income was
$15881. Given this increase in income in both countries, and holding other controls constant,
the overall increase in the steady-state level of the size of government (t¤) in the right-wing
country is 9.41% of GDP, whereas the increase in the left-wing country is 20.32%. This is
a sizeable di¤erence and may explain why Persson and and Tabellini (2004) …nd that the
electoral rule has an impact on the size of government in the 1990s but not in the 1960s:
If majoritarian countries are typically right-wing, and countries with PR electoral rules are
typically left-wing (and Figure 8 is at least suggestive of this), then our model would predict
the data - but of course the mechanism is now straightforward median-voter demand, rather
than due to the constitutional rule itself.
As a further experiment to examine the impact of ideology upon t column (2) drops the
interactive term. Now the ideology variable becomes positive and signi…cant at the 1% level
- giving further credence to the argument that the negative (and insigni…cant) estimate of
the direct e¤ect in column (1) needs to be understood in the context of the presence of the
interaction term. If we take the switch from being archetypally right to archetypally left
(a two-standard deviation change in ideology), then long run impact of this switch is an
increase in the size of government of 5.76%. Nonetheless, this is we argue a crude estimate
of how ideology impacts upon the size of government, because its impact, both theoretically
and empirically, depends strongly on the level of economic development.
Table 2 contains regression results when common time e¤ects are also included. This
table also contains regression results for reduced samples depending on the electoral rule.
20
Whether looking at the full sample or either of the sub-samples the direct impact of income
now becomes insigni…cant as the time e¤ects now do a better job of capturing the common
trend than income itself. The direct impact of ideology is again estimated to be negative,
and in the cases of the PR countries (column 2) this relationship is statistically (if not eco-
nomically) signi…cant. Nonetheless, once again the interactive term is in all three regressions
estimated to be positive and signi…cant at the 1% level, and sizeable. The parameter es-
timates are comparable with the results in table 1, and leftist and rightist regimes would
diverge substantially given representative income increases. A striking conclusion from table
2 is that there are not substantial di¤erences across electoral rules in how government size
responds to income and ideology.
The …nding of a statistically, and economically signi…cant impact of the interaction of
ideology and income holds up in all of our econometric speci…cations. It can explain a
substantial component of the growth, and divergence of government size in OECD countries
over the latter half of the twentieth century.
4 Conclusion
This paper asks how ideology a¤ects the size of government. We de…ne left (right) ideology
as relative (dis-)liking for public provision over private consumption, and in a simple model
where private consumption is a necessity, and public services are luxury goods, then the
impact of ideology increases with the level of income. Using data from party manifestos,
time-varying measures for the median-voter’s ideological position are constructed. These
data have a signi…cant impact upon government size, but most critically, this impact increases
21
with the level of economic development. At 1960s income levels the impact of ideology on
government size is small. At 1990s income levels the impact is large. Our model, and the
ideology data we use can thus explain observations in the 1960s and 1990s and indeed why
there has been such divergence over the OECD countries through this period.
Part of the motivation for the paper was as an investigation of Persson and Tabellini’s
(2004) …nding of signi…cant e¤ects of constitutional rules. We would not argue that the
results presented here are su¢cient to negate their argument. For one thing their data set is
much larger; we are restricted due to the limited availability of ideology data. For another it
is not possible to separate out the non-time-varying constitutional rule from the …xed e¤ect
in panel analysis. Nonetheless, we do feel that ideology casts an important shadow over their
results. Our theory can explain the 1960s observations as well as the 1990s observations,
and it does appear that ideology is correlated with electoral rules, at least in the smaller set
of countries studied here.
22
Appendix - Proof of Proposition 1
part (i)
Di¤erentiating (7) with respect to β gives
∂t¤
∂β=
ϕδ_y + µ (1¡ δm)
ϕδ_y (1 + β)
2[1¡ (1¡ ϕ) δm] .
Proposition 1 requires this expression to be positive. The denominator of this expression
is positive from (9), hence we require ϕδ_y > µ (1¡ δm). From (8) this holds a fortiori if
ϕδnmµ+ [1¡(1¡ϕ)δm]µ
βϕδ
o> µ (1¡ δm). Rearranging, this implies βmµ+[1 ¡ (1¡ ϕ) δm]µ >
βµδm¡βµm. Using again (9) the …rst and second terms on the LHS are respectively greaterthan those on the RHS.
part (ii)
Di¤erentiating (7) with respect to_y gives
∂t¤
∂_y=µ f[1¡ (1 ¡ ϕ) δm] + βϕδmgϕδ
_y2(1 + β) [1¡ (1¡ ϕ) δm]
. (A1)
Given (9) this is positive by inspection. The second derivative is negative, thus establishing
the second part of the proposition:
∂2t¤
∂_y2 = ¡
2µ f[1 ¡ (1¡ ϕ) δm] + βϕδmgϕδ
_y3(1 + β) [1¡ (1 ¡ ϕ) δm]
.
23
part (iii)
Di¤erentiating (A1) with respect to β gives
∂2t¤
∂_y∂β
=µ [δm¡ 1]
ϕδ_y2(1 + β)2 [1¡ (1¡ ϕ) δm]
.
Given (9) the sign of this expression depends on the sign of δm¡ 1.
24
projected government size
0
0.1
0.2
0.3
0.4
0.5
0.6
0 10000 20000 30000 40000 50000
income
size
of t
he S
tate
(t)
archetypal right-wingarchetypal left-wing
Figure 1: Projected government size for archetypal left- and right-wing regimes
Calibration: ϕ = 0.75,m = 2, µ = 2000, δ = 0.8, βL = 0.6, βR = 0.3
25
projected government size
0
0.1
0.2
0.3
0.4
0.5
0.6
0 10000 20000 30000 40000 50000
income
size
of t
he S
tate equal
distributionunequaldistribution
Figure 2: Projected government size and income for equal and unequal income distributions
Calibration: ϕ = 0.75, µ = 2000, δ = 0.8, β = 0.5,mequal = 1.5,munequal = 2.5
26
Figure 4: Size of government (t) and real GDP per capita (1985 PPP). Source: OECDEconomic Outlook and Persson and Tabellini (2003) - from Penn World Tables.
28
Figure 5: Size of government (t) and real GDP per capita (1985 PPP). Source: OECDEconomic Outlook and Persson and Tabellini (2003) - from Penn World Tables.29
Figure 6: Size of government (t) and real GDP per capita (1985 PPP). Source: OECDEconomic Outlook and Persson and Tabellini (2003) - from Penn World Tables.
30
Average median voter ideological position by country, 1945-1998
-30
-25
-20
-15
-10
-5
0
5
10
15
20
Nor
Swe Fin
UK Net
Den Ita Bel
Can
Ger
Fra
Aut
Swit Ire Aus US Ice
country
Deg
ree
of c
onse
rvat
ism
Figure 7: Average median voter ideological position by country 1945-1998. Source: Budgeet al (2001).
31
Average median voter ideology in Majoritarian and PR countries
-20
-15
-10
-5
0
5
10
1950
1952
1954
1956
1958
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
year
degr
ee o
f con
serv
atis
m
MajoritarianPR
Figure 8: Average Median Voter ideology in majoritarian and PR countries. Author’s cal-culations using data from Budge et al (2001).
32
average government size in the 90s versus average median voter ideology 1945-98
Nor
Swe
Fin
UK
Den
Nld
Ita
BelFra
Aut
Ire
Ger
Can
Ice
Che
AusUSA
y = 127.42x + 47.439R2 = 0.4905
30.0
35.0
40.0
45.0
50.0
55.0
60.0
65.0
70.0
-0.1 -0.05 0 0.05 0.1 0.15
ideology (left-wing scale)
size
of g
over
nmen
t
Figure 9: Average government size in the 90s versus average median voter ideology 1945-1998.
33
(1a) (1b) (2a) (2b)
L.Outlays0.883
(0.017)¤¤¤
0.892
(0.017)¤¤¤
Y0.157
(0.048)¤¤¤
1.345
(0.351)¤¤¤
0.150
(0.048)¤¤¤
Ideo¡2.772(2.552)
¡23.68(21.16)
2.663
(1.209)¤¤¤
24.351
(12.35)¤¤
Y-I0.548
(0.228)¤¤¤
4.685
(1.923)¤¤¤
Sample OECD OECD OECD OECD
Number of Observations 625 625 625 625
Number of Countries 17 17 17 17
R2 (within) 0.95 0.95
Table 1: Dynamic panel estimation with …xed e¤ects
Notes: Panel regressions of Government Outlays as a percentage share of GDP including
PROP1564, PROP65, TRADE, YGAP, OIL_EX, OIL_IM as control variables. OECD
comprises the entire data set. L.Outlays is the lagged dependent variable. Y is income per
capita in $000s of 1985 0prices (PPP), ideo is the ideology variable described in the text,
Y-I is the interaction (product) of Y and ideo. Standard errors are shown in parentheses.
Columns (1b) and (2b) contain ‘long-run’ parameter estimates - see the text for details on
how these are estimated.
34
(1) (2) (3)
L.Outlays0.913
(0.017)¤¤¤
0.913
(0.023)¤¤¤
0.892
(0.017)¤¤¤
Y¡0.091(0.109)
¡0.131(0.154)
0.308
(0.186)
Ideo¡4.071(2.153)¤
¡7.469(2.871)¤¤¤
¡4.555(3.872)
Y-I0.471
(0.191)¤¤¤
0.806
(0.275)¤¤¤
0.572
(0.298)¤¤¤
Sample OECD PR Maj
Number of Observations 625 437 188
Number of Countries 17 13 5
R2 (within) 0.97 0.97 0.98
Table 2: Dynamic panel estimation with …xed e¤ects and time e¤ects
Notes: As for table 1. PR comprises the subset of observations with Proportional Represen-
tation electoral rules, Maj the subset of observations with Majoritarian electoral rules
35
References
Acemoglu, Daron (2005). Constitutions, politics and economics: a review essay on Persson
and Tabellini’s "The economic e¤ects of constitutions". Journal of Economic Literature
43 1025-1048.
Alesina, Alberto, Edward Glaeser and Bruce Sacerdote (2001). Why doesn’t the US have a
European-style welfare state? Brookings Papers on Economic Activity 2, 187-277.
Alesina, Alberto and George-Marios Angeletos (2005). Fairness and Redistribution: US vs
Europe. American Economic Review 95, 960-80.
Baumol, William J. (1967). Macroeconomics of unbalanced growth: The anatomy of an
urban crisis. American Economic Review 57, 415-426.
Budge, Ian, Hans-Dieter Klingemann, Andrea Volkens, Judith Bara and Eric Tanenbaum
(2001) Mapping policy preferences: estimates for parties, electors and governments
1945-1998. OUP, Oxford.
Gabel, Matthew and Simon Hix. (2005) Reconsidering the Economic E¤ects of Constitutions.
Unpublished research paper.
Henrekson, Magnus and Lybeck, Johan A. (1988) Chapter 1 Editors’ Introduction and Sum-
mary in Explaining the growth of Government, ed.s Johan A. Lybeck and Magnus Hen-
rekson. Elsevier Science. Amsterdam, The Netherlands.
Holsey, Cheryl M. and Thomas E. Borcherding (1997) Why does government’s share grow?
36
An assessment of the recent literature on the US. in Perspectives on Public Choice: A
Handbook. ed. Dennis C. Mueller. Cambridge University Press, New York, 562-591.
Kau, J. B. and P. H. Rubin (2002) The growth of government: sources and limits. Public
Choice 113, 389-402.
Kim, Hee-Min and Richard C, Fording (2001) Extending Party Estimates to Governments
and Electors. Chapter 8 in Budge et al (2004).
Lizzeri, Alessandro and Nicola Persico (2001) The provision of public goods under alternative
electoral incentives. American Economic Review 91(1) 225-45.
Lybeck, Johan A. (1988) Comparing government growth rates: the non-institutional vs the
institutional approach, in Explaining the growth of Government, ed.s Johan A. Lybeck
and Magnus Henrekson. Elsevier Science. Amsterdam, The Netherlands.
Meltzer, Allan H, and Richard, Scott F. (1981). A rational theory of the size of government.
Journal of Political Economy 89, 914-927.
Meltzer, Allan H, and Richard, Scott F. (1983). Tests of a rational theory of the size of
government. Public Choice 41, 403-418
Milesi-Ferretti, Gian-Maria, Roberto Perotti and Massimo Rostagno (2002). Electoral sys-
tems and public spending. Quarterly Journal of Economics 117, 609-657.
Peacock, Alan, and Alex Scott (2000). The curious attraction ofWagner’s law. Public Choice
102, 1-17.
37
Persson, Torsten, Gerard Roland andGuido Tabellini (2000)Comparative politics and public
…nance. Journal of Political Economy 117(2) 609-57.
Persson, Torsten and Guido Tabellini (1999) The size and scope of government: comparative
politics with rational politicians. European Economic Review 43(4-6) 699-735.
Persson, Torsten and Guido Tabellini (2000). Political economics: Explaining economic pol-
icy. MIT Press, Cambridge, MA.
Persson, Torsten and Guido Tabellini (2003). The economic e¤ects of constitutions. MIT
Press, Cambridge, MA.
Persson, Torsten andGuido Tabellini (2004). Constitutional rules and …scal policy outcomes.
American Economic Review 94(1), 25-45.
Wagner, Adolph (1893) Grundlegung der Politischen Oekonomie, 3rd ed. Leipzig: C. F.
Winter.
38